CN110399785A - A kind of detection method that the leaf based on deep learning and traditional algorithm blocks - Google Patents
A kind of detection method that the leaf based on deep learning and traditional algorithm blocks Download PDFInfo
- Publication number
- CN110399785A CN110399785A CN201910466457.4A CN201910466457A CN110399785A CN 110399785 A CN110399785 A CN 110399785A CN 201910466457 A CN201910466457 A CN 201910466457A CN 110399785 A CN110399785 A CN 110399785A
- Authority
- CN
- China
- Prior art keywords
- image
- leaf
- blocks
- neural networks
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 238000012360 testing method Methods 0.000 claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 46
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 33
- 238000013145 classification model Methods 0.000 claims abstract description 21
- 230000006870 function Effects 0.000 claims abstract description 11
- 230000003796 beauty Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 abstract description 8
- 238000012545 processing Methods 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000000034 method Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000033772 system development Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Tourism & Hospitality (AREA)
- Computing Systems (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Multimedia (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to technical field of image processing, are specifically related to a kind of detection method that the leaf based on deep learning and traditional algorithm blocks.Training set is input in convolutional neural networks model, convolutional neural networks model is trained, until terminating the training to convolutional neural networks model when the loss function of convolutional neural networks model meets setting condition, convolutional neural networks model after training is optimal classification model.The image for using traditional algorithm to judge that whether test set image blocks for no leaf does the detection of a step using deep learning algorithm if traditional algorithm detects that image is to have leaf to block to image.It is capable of detecting when whether the monitor camera is blocked by leaf, staff's timely learning monitor camera is facilitated to block this case by leaf, to adjust the position of the monitor camera, prevent the missing of monitoring scene, reduce the influence to public safety.
Description
Technical field
The present invention relates to technical field of image processing, are specifically related to a kind of leaf based on deep learning and traditional algorithm
The detection method blocked.
Background technique
With the rapid growth of video monitoring service, there are the problem of be also gradually exposed, how at the first time
It was found that front end monitor camera failure, improves the maintenance work efficiency of video monitoring system, it is that video monitoring system development can not
Or scarce a part.
It is very common one of failure that leaf, which blocks, in monitor camera failure, since monitor camera position is constant, and
Trees near it can constantly grow with the time, arrive specific season or time, the monitor camera of original normal mounting
It can be blocked by leaf.If the monitor camera blocked by leaf cannot be detected, the missing of monitoring scene will cause, give public peace
It affects entirely.
Summary of the invention
In order to solve the above technical problems, blocked the present invention provides a kind of leaf based on deep learning and traditional algorithm
Detection method is capable of detecting when the monitor camera blocked by leaf, prevents the missing of monitoring scene, reduces to public safety
It influences.
To achieve the above object, the invention adopts the following technical scheme:
A kind of detection method that the leaf based on deep learning and traditional algorithm blocks, includes the following steps:
Training set is input in convolutional neural networks model by S1, convolutional neural networks, that is, deep learning algorithm, training set
In include image that leaf blocks and without the image that leaf blocks, the image tag for having leaf to block is 1, what no leaf blocked
Image tag is 0；
S2 is trained the convolutional neural networks model in step S1, until the loss letter of convolutional neural networks model
When number meets setting condition, the training to convolutional neural networks model is terminated, convolutional neural networks model after training is most
Good disaggregated model；
S3, the image for using traditional algorithm to judge that whether test set image blocks for no leaf, i.e. extraction test set image
Feature, feature include at least color characteristic and textural characteristics, if the color characteristic P of test set image1Less than first threshold and
Textural characteristics P2Greater than second threshold, then otherwise the image that the test set image blocks for no leaf carries out step S4；
Test set image is input in the optimal classification model in step S2 by S4, if optimal classification model output 1,
The test set image is the image for having leaf to block；If optimal classification model output 0, which blocks for no leaf
Image.
Further, the structure of convolutional neural networks model is first the-the first pond of convolutional layer the-the second convolution of layer in step S1
The-the second full articulamentum-softmax of the-the first full articulamentum-the second of pond layer of layer；Training set is input to convolution mind in step S1
It is the first convolutional layer that training set is input to convolutional neural networks model through network model；It is in step S4 that test set image is defeated
Entering to optimal classification model is the first convolutional layer that test set image is input to optimal classification model, optimal classification model
Softmax output 1, then the test set image is the image for having leaf to block, if the softmax output 0 of optimal classification model,
The image that the test set image blocks for no leaf.
Further, specific step is as follows by step S3:
The image of test set is transformed into hsv color space from RGB color, wherein R is beauty chrominance channel, and G is green
Color Channel, B are blue Color Channel, and H is tone channel, and S is saturation degree channel, and V is luminance channel；Extract tone channel H's
Value, calculates the ratio that the pixel for meeting and imposing a condition accounts for the test set image all pixels point, which obtains step S3
In color characteristic P1；
Textural characteristics in step S3 are corner feature, and the picture of test set image corner point is calculated using Corner Detection Algorithm
Vegetarian refreshments accounts for the ratio of the test set image all pixels point, which is the textural characteristics P obtained in step S32。
It is further preferred that pre-processing before step S1 to acquired image, the image after pre-processing is made
For training set and test set, the specific steps are as follows:
S00 adjusts the size of each acquired image, so that the size of all acquired images is identical, adjustment is big
Image after small is set as original image；
S01, adjusts the contrast of original image, and the original image after adjustment contrast is set as JPG1；Adjust the bright of original image
Degree, the original image after adjustment brightness are set as JPG2；
S02, anticlockwise original image, JPG1 and JPG2, original image, JPG1 and JPG2 after anticlockwise are set to respectively
The anticlockwise image of the anticlockwise image of original image, the anticlockwise image of JPG1 and JPG2；Respectively right rotation original image, JPG1 and
JPG2, original image, JPG1 and JPG2 after right rotation are set to the right rotation image of the right rotation image of original image, JPG1
With the right rotation image of JPG2；Training set and test set include original image, JPG1, JPG2, original image anticlockwise image,
The anticlockwise image of JPG1, the anticlockwise image of JPG2, the right rotation image of original image, the right rotation image of JPG1 and JPG2
Right rotation image.
It is further preferred that when the setting condition that the loss function of step S2 meets is loss function convergence, volume at this time
Product neural network model is optimal classification model.
It is further preferred that the Corner Detection Algorithm in step S32 is canny operator.
It is further preferred that the setting condition in step S3 is 30 < H < 80.
It is further preferred that the leaf type in training set image includes at least two classes；At least two width in training set image
The leaf depth of image is different；The acquisition scene of training set image includes at least two scenes；Acquire the weather of training set image
Including at least two kinds of weather；The brightness value of training set image includes at least two.
Beneficial effects of the present invention are as follows:
(1) video image one that detection method of the invention only needs monitor camera to take is opened, by detecting the image
Whether blocked by leaf, and then be capable of detecting when whether the monitor camera is blocked by leaf, facilitates staff's timely learning
Monitor camera blocks this case by leaf, to adjust the position of the monitor camera, prevents the missing of monitoring scene, drops
The low influence to public safety.
(2) detection method of the invention combines deep learning algorithm with traditional algorithm, keeps the apparent color of leaf special
Corner feature of seeking peace gets application, and solves and deep learning or traditional algorithm progress leaf occlusion detection false detection rate height is used alone
The problem of, improve the accuracy rate of detection.
(3) deep learning algorithm is combined with traditional algorithm, when detecting between be basically unchanged, but significantly improve detection
Accuracy rate.
(4) structure of convolutional neural networks model is designed to two convolutional layers, two pond layers, two full articulamentums, one
A recurrence layer, can be improved the accuracy rate of detection.
(5) channel H in hsv color space can intuitively indicate color, it is easier to set the tone section of leaf.
(6) training set image and test set image have carried out pretreatment operation, have expanded sample database.
(7) training set image is blocked comprising different leaf types, different coverage extents, different far and near leaves, acquisition
Scene diversification, weather diversification, brightness diversification, so that algorithm generalization ability is strong.
Detailed description of the invention
Fig. 1 is the schematic diagram of detection method of the invention；
Fig. 2 is convolutional neural networks illustraton of model of the invention；
Fig. 3 a-3d is the image that no leaf of the invention blocks；
Fig. 4 a and Fig. 4 b have the image that leaf blocks for of the invention.
Specific embodiment
With reference to embodiments and Figure of description, the technical solution in the present invention is clearly and completely described.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
His embodiment, shall fall within the protection scope of the present invention.
Embodiment 1
A kind of detection method that the leaf based on deep learning and traditional algorithm blocks, includes the following steps:
S1 adjusts the size of collected each test set image and each training set image, so that all images is big
Small identical, the size of all images is 64*64 in the present embodiment；Each test set image and each training set image are carried out
Anticlockwise and right rotation, each test set image and each training set image can derive two images, increase sample size；
The contrast for adjusting each test set image and each training set image increases so that the same image has multiple contrasts
Sample size adjusts the brightness of each test set image and each training set image, for increasing test set image and training set figure
The sample size of picture.
Specific step is as follows:
S10 adjusts the size of each acquired image, so that the size of all acquired images is identical, adjustment is big
Image after small is set as original image；
S11, adjusts the contrast of original image, and the original image after adjustment contrast is set as JPG1；Adjust the bright of original image
Degree, the original image after adjustment brightness are set as JPG2；
S12, anticlockwise original image, JPG1 and JPG2, original image, JPG1 and JPG2 after anticlockwise are set to respectively
The anticlockwise image of the anticlockwise image of original image, the anticlockwise image of JPG1 and JPG2；Respectively right rotation original image, JPG1 and
JPG2, original image, JPG1 and JPG2 after right rotation are set to the right rotation image of the right rotation image of original image, JPG1
With the right rotation image of JPG2；Training set and test set include original image, JPG1, JPG2, original image anticlockwise image,
The anticlockwise image of JPG1, the anticlockwise image of JPG2, the right rotation image of original image, the right rotation image of JPG1 and JPG2
Right rotation image
In the present embodiment, as shown in figures 4 a and 4b, training set image include different leaf types, different coverage extent,
Different far and near leaves block, scene diversification, weather diversification, the brightness diversification of acquisition, so that the extensive energy of detection method
Power is strong, that is, increases the application range of detection method.
As illustrated in figs. 3 a-3d, the image that no leaf blocks, the monitoring scene that video camera takes are complete.
It is more as far as possible to block the leaf type for requiring to block in sample, includes the most of shade tree types in city；It is required that having big
Area, which is blocked, different degrees of block such as blocks with small area；The different distances such as block and closely block at a distance it is required that having
It blocks；Sample scene requirement includes indoor and outdoor, urban road, rural field；Unshielding sample requires have lawn, bushes etc.
It may be also easy to produce the sample of erroneous detection, there is tree but do not form the sample blocked, and the normal sample that do not set；Sample needs
The sample of the different times such as fine day rainy day, dusk on daytime；The sample of training set is clear as far as possible, and ambiguous two are not present on manual sort
Can the case where.
S2, as shown in Figure 1, training set is input in convolutional neural networks model, convolutional neural networks, that is, deep learning
Algorithm includes image that leaf blocks and without the image that leaf blocks in training set, and the image tag for having leaf to block is 1,
The image tag that no leaf blocks is 0；
S3 is trained the convolutional neural networks model in step S2, and every iteration is primary, is updated using gradient descent method
Loss function, until terminating when the loss function of convolutional neural networks model meets setting condition to convolutional neural networks model
Training, convolutional neural networks model after training is optimal classification model；
In the present embodiment, as shown in Fig. 2, convolutional neural networks modelling is at two convolutional layers, two pond layers, two
Full articulamentum, a recurrence layer.It is complete to be followed successively by first the-the first pond of convolutional layer the-the second pond of the-the second convolutional layer of layer layer-the first
Articulamentum-the second full articulamentum-softmax, softmax, that is, differentiation layer, the purpose of softmax are by the probability normalizing of result
Change, then exports the result of maximum probability.First convolutional layer uses the convolution kernel of 64 3*3*3, and second layer convolutional layer uses 16
The convolution kernel of a 3*3*64.The time required to reducing training, trained sample is adjusted to 64*64 size.
In the present embodiment, impose a condition for loss function convergence when, i.e., loss function only change near some value, at this time
Convolutional neural networks model be optimal classification model.The training time is too long in order to prevent or loss function misses optimum point,
Initial value can be set to learning rate.
S4, the image for using traditional algorithm to judge that whether test set image blocks for no leaf, i.e. extraction test set image
Feature, feature include at least color characteristic and textural characteristics, if the color characteristic P of test set image1Less than first threshold and
Textural characteristics P2Greater than second threshold, then otherwise the image that the test set image blocks for no leaf carries out step S5.
In the present embodiment, extract color characteristic the step of: the image of test set is transformed into HSV face from RGB color
The colour space, wherein R is beauty chrominance channel, and G is green color channel, and B is blue Color Channel, and H is tone channel, and S is logical for saturation degree
Road, V are luminance channel.The value of tone channel H is extracted, the pixel for calculating 30 < H < 80 accounts for the test set image all pixels
The ratio of point, i.e. color characteristic P1.In the present embodiment, H can go 31,50,75.
In the present embodiment, texture feature extraction is the corner feature for extracting image, is calculated and is tested using Corner Detection Algorithm
The pixel of collection image corner point accounts for the ratio of the test set image all pixels point, i.e. textural characteristics P2.In the present embodiment
Corner Detection Algorithm is canny operator.
In the present embodiment, first threshold 0.5, second threshold 0.2.
Test set image is input in the optimal classification model in step S3 by S5, and optimal classification model output 1 then should
Test set image is the image for having leaf to block；Optimal classification model output 0, the then figure that the test set image blocks for no leaf
Picture.
Embodiment 2
On the basis of embodiment 1, table 1 is test set image, sample number unit: is opened
Table 1
In table 1: the image that no leaf blocks is the image blocked without leaf for including city, the more scenes in rural area；
Number/reality of verification and measurement ratio=the be detected as image that leaf blocks is the number for the image for having leaf to block
Accuracy rate=being detected as leaf to block and be actually the number for the image for having leaf to block/has been detected as leaf screening
The number of the image of gear.
Claims (8)
1. a kind of detection method that the leaf based on deep learning and traditional algorithm blocks, which comprises the steps of:
Training set is input in convolutional neural networks model by S1, convolutional neural networks, that is, deep learning algorithm, is wrapped in training set
The image that blocks containing leaf and without the image that leaf blocks, the image tag for having leaf to block is 1, the image that no leaf blocks
Label is 0；
S2 is trained the convolutional neural networks model in step S1, until the loss function of convolutional neural networks model is full
When foot imposes a condition, the training to convolutional neural networks model is terminated, convolutional neural networks model after training is best point
Class model；
S3, the image for using traditional algorithm to judge that whether test set image blocks for no leaf extract the spy of test set image
Sign, feature includes at least color characteristic and textural characteristics, if the color characteristic P of test set image1Less than first threshold and texture
Feature P2Greater than second threshold, then otherwise the image that the test set image blocks for no leaf carries out step S4；
Test set image is input in the optimal classification model in step S2 by S4, if optimal classification model output 1, the survey
Examination collection image is the image for having leaf to block；If optimal classification model output 0, the figure which blocks for no leaf
Picture.
2. the detection method that leaf as described in claim 1 blocks, it is characterised in that: convolutional neural networks model in step S1
Structure to be that the full articulamentum-the second of first the-the first pond of convolutional layer the-the second pond of the-the second convolutional layer of layer layer-the first is complete connect
Layer-softmax；It is that training set is input to convolutional neural networks that training set, which is input to convolutional neural networks model, in step S1
First convolutional layer of model；It is to be input to test set image most that test set image, which is input to optimal classification model, in step S4
First convolutional layer of good disaggregated model, the softmax output 1 of optimal classification model, then the test set image is to have leaf to block
Image, if optimal classification model softmax output 0, the image which blocks for no leaf.
3. the detection method that leaf as claimed in claim 1 or 2 blocks, which is characterized in that specific step is as follows by step S3:
The image of test set is transformed into hsv color space from RGB color, wherein R is beauty chrominance channel, and G is green color
Channel, B are blue Color Channel, and H is tone channel, and S is saturation degree channel, and V is luminance channel；Extract the value of tone channel H, meter
The ratio that the pixel for meeting and imposing a condition accounts for the test set image all pixels point is calculated, which is the face obtained in step S3
Color characteristic P1；
Textural characteristics in step S3 are corner feature, and the pixel of test set image corner point is calculated using Corner Detection Algorithm
The ratio of the test set image all pixels point is accounted for, which is the textural characteristics P obtained in step S32。
4. the detection method that leaf as claimed in claim 3 blocks, which is characterized in that before step S1, to collected
Image is pre-processed, and the image after pre-processing is as training set and test set, the specific steps are as follows:
S00 adjusts the size of each acquired image, so that the size of all acquired images is identical, is sized it
Image afterwards is set as original image；
S01, adjusts the contrast of original image, and the original image after adjustment contrast is set as JPG1；The brightness of original image is adjusted, is adjusted
Original image after whole brightness is set as JPG2；
S02, anticlockwise original image, JPG1 and JPG2, original image, JPG1 and JPG2 after anticlockwise are set to original image respectively
The anticlockwise image of the anticlockwise image of picture, the anticlockwise image of JPG1 and JPG2；Respectively right rotation original image, JPG1 and
JPG2, original image, JPG1 and JPG2 after right rotation are set to the right rotation image of the right rotation image of original image, JPG1
With the right rotation image of JPG2；Training set and test set include original image, JPG1, JPG2, original image anticlockwise image,
The anticlockwise image of JPG1, the anticlockwise image of JPG2, the right rotation image of original image, the right rotation image of JPG1 and JPG2
Right rotation image.
5. the detection method that leaf as claimed in claim 4 blocks, which is characterized in that the loss function of step S2 met sets
When fixed condition is that loss function is restrained, convolutional neural networks model at this time is optimal classification model.
6. the detection method that leaf as claimed in claim 3 blocks, which is characterized in that the Corner Detection Algorithm in step S32
For canny operator.
7. the detection method that leaf as claimed in claim 3 blocks, which is characterized in that the setting condition in step S3 is 30 <
H < 80.
8. the detection method that leaf as claimed in claim 1 or 2 blocks, which is characterized in that the leaf kind in training set image
Class includes at least two classes；The leaf depth of at least two images is different in training set image；The acquisition scene of training set image is extremely
It less include two scenes；The weather for acquiring training set image includes at least two kinds of weather；The brightness value of training set image at least wraps
Include two.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910466457.4A CN110399785B (en) | 2019-05-31 | 2019-05-31 | Method for detecting leaf occlusion based on deep learning and traditional algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910466457.4A CN110399785B (en) | 2019-05-31 | 2019-05-31 | Method for detecting leaf occlusion based on deep learning and traditional algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110399785A true CN110399785A (en) | 2019-11-01 |
| CN110399785B CN110399785B (en) | 2021-04-06 |
Family
ID=68323058
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910466457.4A Active CN110399785B (en) | 2019-05-31 | 2019-05-31 | Method for detecting leaf occlusion based on deep learning and traditional algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110399785B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046956A (en) * | 2019-12-13 | 2020-04-21 | 苏州科达科技股份有限公司 | Occlusion image detection method and device, electronic equipment and storage medium |
| CN111144280A (en) * | 2019-12-25 | 2020-05-12 | 苏州奥易克斯汽车电子有限公司 | Monitoring video leaf occlusion detection method |
| CN114713531A (en) * | 2022-04-25 | 2022-07-08 | 深圳智优停科技有限公司 | Convolutional neural network training method, lens contamination type judgment method, lens wiper control method, storage medium and shooting system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102111532A (en) * | 2010-05-27 | 2011-06-29 | 周渝斌 | Camera lens occlusion detecting system and method |
| CN103139547A (en) * | 2013-02-25 | 2013-06-05 | 昆山南邮智能科技有限公司 | Method of judging shielding state of pick-up lens based on video image signal |
| CN104504707A (en) * | 2014-12-26 | 2015-04-08 | 西安理工大学 | Foreign matter blocking detecting method of monitoring camera video images |
| CN108805042A (en) * | 2018-05-25 | 2018-11-13 | 武汉东智科技股份有限公司 | The detection method that road area monitor video is blocked by leaf |
-
2019
- 2019-05-31 CN CN201910466457.4A patent/CN110399785B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102111532A (en) * | 2010-05-27 | 2011-06-29 | 周渝斌 | Camera lens occlusion detecting system and method |
| CN103139547A (en) * | 2013-02-25 | 2013-06-05 | 昆山南邮智能科技有限公司 | Method of judging shielding state of pick-up lens based on video image signal |
| CN104504707A (en) * | 2014-12-26 | 2015-04-08 | 西安理工大学 | Foreign matter blocking detecting method of monitoring camera video images |
| CN108805042A (en) * | 2018-05-25 | 2018-11-13 | 武汉东智科技股份有限公司 | The detection method that road area monitor video is blocked by leaf |
Non-Patent Citations (3)
| Title |
|---|
| 王宝君 等: "基于角点的监控摄像头干扰检测", 《计算机应用与软件》 * |
| 袁渊 等: "基于颜色与纹理特征的安防视频遮挡树叶检测", 《计算机工程与设计》 * |
| 邬美银 等: "基于深度学习的监控视频树叶遮挡检测", 《武汉科技大学学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046956A (en) * | 2019-12-13 | 2020-04-21 | 苏州科达科技股份有限公司 | Occlusion image detection method and device, electronic equipment and storage medium |
| WO2021114866A1 (en) * | 2019-12-13 | 2021-06-17 | 苏州科达科技股份有限公司 | Method and apparatus for detecting occluded image, electronic device, and storage medium |
| CN111144280A (en) * | 2019-12-25 | 2020-05-12 | 苏州奥易克斯汽车电子有限公司 | Monitoring video leaf occlusion detection method |
| CN114713531A (en) * | 2022-04-25 | 2022-07-08 | 深圳智优停科技有限公司 | Convolutional neural network training method, lens contamination type judgment method, lens wiper control method, storage medium and shooting system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110399785B (en) | 2021-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570486B (en) | Filtered target tracking is closed based on the nuclear phase of Fusion Features and Bayes's classification | |
| CN103578119B (en) | Target detection method in Codebook dynamic scene based on superpixels | |
| CN110399785A (en) | A kind of detection method that the leaf based on deep learning and traditional algorithm blocks | |
| CN107578390A (en) | A kind of method and device that image white balance correction is carried out using neutral net | |
| CN110479636B (en) | Method and device for automatically sorting tobacco leaves based on neural network | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN106296695A (en) | Adaptive threshold natural target image based on significance segmentation extraction algorithm | |
| CN110427922A (en) | One kind is based on machine vision and convolutional neural networks pest and disease damage identifying system and method | |
| CN106845497B (en) | Corn early-stage image drought identification method based on multi-feature fusion | |
| CN106412619A (en) | HSV color histogram and DCT perceptual hash based lens boundary detection method | |
| CN103927520A (en) | Method for detecting human face under backlighting environment | |
| CN106373096B (en) | A kind of shadow removing method of multiple features Weight number adaptively | |
| CN106973240A (en) | Realize the digital camera imaging method that high dynamic range images high definition is shown | |
| CN110070526A (en) | Defect inspection method based on the prediction of deep neural network temperature figure | |
| CN108428239A (en) | Intelligent grass-removing Boundary Recognition method based on image texture characteristic extraction | |
| CN110490252A (en) | A kind of occupancy detection method and system based on deep learning | |
| CN106845366B (en) | Sugarcane coverage automatic detection method based on image | |
| CN108564021B (en) | Method for extracting vegetation coverage of desert based on digital photos | |
| CN112257702A (en) | Crop disease identification method based on incremental learning | |
| CN109726669A (en) | Pedestrian identifies data creation method again under different illumination conditions based on confrontation network | |
| CN109949353A (en) | A kind of low-light (level) image natural sense colorization method | |
| CN107221070A (en) | A kind of bill anti-counterfeit discrimination method recognized based on master pattern fluorescent characteristics | |
| CN107346429A (en) | The color automatic identification and sorting technique of a kind of polycrystalline battery picture | |
| CN108898098A (en) | Early stage video smoke detection method based on monitor supervision platform | |
| CN109086823A (en) | A kind of wheat scab disease tassel yield method for automatically counting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |